kb-office-remediation

Pass

Audited by Gen Agent Trust Hub on Sep 26, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill provides patterns for processing external Office documents (Word, Excel, PowerPoint). While the skill itself is static documentation, agents implementing these patterns ingest untrusted file content which presents a surface for indirect prompt injection if the documents contain instructions targeted at the agent.
  • Ingestion points: Processes .docx, .xlsx, and .pptx files via libraries and COM automation.
  • Boundary markers: None provided in the reference snippets.
  • Capability inventory: File system read/write, COM automation (PowerShell), and shell execution (Bash snippets for unzip/zip/sed).
  • Sanitization: No explicit sanitization of document content is demonstrated in the remediation patterns.
  • [COMMAND_EXECUTION]: The skill includes code snippets that execute shell commands and COM automation.
  • Evidence: PowerShell snippets use New-Object -ComObject to interact with Word, Excel, and PowerPoint applications.
  • Evidence: Bash snippets use unzip, sed, and zip for direct OOXML manipulation.
  • Context: These operations are directly related to the skill's primary purpose of Office document remediation and are provided as standard technical examples.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 26, 2026, 05:59 AM